Authors:
Reinaldo Silva Fortes
1
;
Alan R. R. de Freitas
2
and
Marcos André Gonçalves
3
Affiliations:
1
Universidade Federal de Minas Gerais and Universidade Federal de Ouro Preto, Brazil
;
2
Universidade Federal de Ouro Preto, Brazil
;
3
Universidade Federal de Minas Gerais, Brazil
Keyword(s):
Recommender Systems, Information Filtering, Hybrid Filtering, Collaborative Filtering, Content-based Filtering, Meta-feature.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Data Mining
;
Databases and Information Systems Integration
;
Enterprise Information Systems
;
Knowledge Discovery and Information Retrieval
;
Knowledge-Based Systems
;
Sensor Networks
;
Signal Processing
;
Soft Computing
;
Symbolic Systems
;
User Profiling and Recommender Systems
Abstract:
Recommender Systems (RS) may behave differently depending on the characteristics of the input data, encouraging
the development of Hybrid Filtering (HF). There are few works in the literature that explicitly
characterize aspects of the input data and how they can lead to better HF solutions. Such work is limited to
the scope of combination of Collaborative Filtering (CF) solutions, using only rating prediction accuracy as
an evaluation criterion. However, it is known that RS also need to consider other evaluation criteria, such as
novelty and diversity, and that HF involving more than one approach can lead to more effective solutions. In
this work, we begin to explore this under-investigated area, by evaluating different HF strategies involving
CF and Content-Based (CB) approaches, using a variety of data characteristics as extra input data, as well
as different evaluation criteria. We found that the use of data characteristics in HF proved to be useful when
considering different eva
luation criteria. This occurs in spite of the fact that the experimented methods aim at
minimizing only the rating prediction errors, without considering other criteria.
(More)